Self-Driving Car Engineer Nanodegree¶
Deep Learning¶
Project: Build a Traffic Sign Recognition Classifier¶
In this notebook, a template is provided for you to implement your functionality in stages which is required to successfully complete this project. If additional code is required that cannot be included in the notebook, be sure that the Python code is successfully imported and included in your submission, if necessary. Sections that begin with 'Implementation' in the header indicate where you should begin your implementation for your project. Note that some sections of implementation are optional, and will be marked with 'Optional' in the header.
In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.
Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. In addition, Markdown cells can be edited by typically double-clicking the cell to enter edit mode.
Step 0: Load The Data¶
# Load pickled data
import pickle
# TODO: Fill this in based on where you saved the training and testing data
training_file = '/home/ros/SDC/CarND-Traffic-Sign-Classifier-Project/train.p'
testing_file = '/home/ros/SDC/CarND-Traffic-Sign-Classifier-Project/test.p'
with open(training_file, mode='rb') as f:
train = pickle.load(f)
with open(testing_file, mode='rb') as f:
test = pickle.load(f)
x_train, y_train = train['features'], train['labels']
x_test, y_test = test['features'], test['labels']
n_classes = len(set(train['labels']))
x_train_split=x_train
y_train_split=y_train
print("Training Set Total: {} samples".format(len(x_train)))
Step 1: Dataset Summary & Exploration¶
The pickled data is a dictionary with 4 key/value pairs:
'features'the images pixel values, (width, height, channels)'labels'is a 2D array containing the label/class id of the traffic sign. The filesignnames.csvcontains id -> name mappings for each id.'sizes'is a list containing tuples, (width, height) representing the the original width and height the image.'coords'is a list containing tuples, (x1, y1, x2, y2) representing coordinates of a bounding box around the sign in the image. THESE COORDINATES ASSUME THE ORIGINAL IMAGE. THE PICKLED DATA CONTAINS RESIZED VERSIONS (32 by 32) OF THESE IMAGES
Complete the basic data summary below.
#The traffic sign data set dosen't come with data split therefore
#import sklearn model for splitting data into learninig traininig and validation
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix
#take 20% of data of the training set for testing(validation)
n_test = x_test.shape[0]
x_train_split=x_train
y_train_split=y_train
x_train,x_validation,y_train, y_validation = train_test_split(x_train_split, y_train_split, test_size=0.2, random_state=0)
x_train = x_train.astype('float32')
x_validation = x_validation.astype('float32')
x_test= x_test.astype('float32')
x_train = x_train / 255 - 0.5
x_validation = x_validation / 255 - 0.5
x_test = x_test / 255 - 0.5
print("Updated image shape {}".format (x_train[0].shape))
print()
print("Training Set: {} samples".format(len(x_train)))
print("Validation Set: {} samples".format(len(x_validation)))
print("Test Set: {} samples".format(len(x_test)))
print("Number of classes =", n_classes)
Visualize the German Traffic Signs Dataset using the pickled file(s). This is open ended, suggestions include: plotting traffic sign images, plotting the count of each sign, etc.
The Matplotlib examples and gallery pages are a great resource for doing visualizations in Python.
NOTE: It's recommended you start with something simple first. If you wish to do more, come back to it after you've completed the rest of the sections.
### Data exploration visualization goes here.
### Feel free to use as many code cells as needed.
#Visualize one sample and the corresponding class to verify dataare consistant
import random
import csv
import math
from tqdm import tqdm
#import cv2
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
for i in range(0, n_classes):
image = train['features'][train['labels'] == i][5]
plt.figure(figsize=(3,3))
plt.imshow(image)
plt.show()
print(i)
#Plot a histogram of data in order to see the frequency of each sign/class
#Make sure the excell file is included in the director in order to read the meanining of each class, will look cool!
sign_names = {}
n_samples = []
#Open excel file and read its contents for the sign labels(meaning)
with open('/home/ros/SDC/CarND-Traffic-Sign-Classifier-Project/signnames.csv', 'r') as file:
read_file = csv.DictReader(file)
for i in read_file:
sign_names.setdefault(i['ClassId'],[]).append(i['SignName'])
#Itirate through the hist
for i in range(0, n_classes):
n_samples.append(len(train['features'][train['labels'] == i]))
print("Count of", sign_names[str(i)], "sign is", n_samples[i], "order is", i)
plt.figure(1)
n_class_list, _, _ = plt.hist(y_train, n_classes)
plt.title("Samples Histogram (Training data)")
plt.xlabel("Classes")
plt.ylabel("Frequency (Classes)")
plt.grid(True)
plt.figure(2)
plt.hist(y_test, n_classes)
plt.title("Samples Histogram (Testing data)")
plt.xlabel("Classes")
plt.ylabel("Frequency (Classes)")
plt.grid(True)
plt.show()
Step 2: Design and Test a Model Architecture¶
Design and implement a deep learning model that learns to recognize traffic signs. Train and test your model on the German Traffic Sign Dataset.
There are various aspects to consider when thinking about this problem:
- Neural network architecture
- Play around preprocessing techniques (normalization, rgb to grayscale, etc)
- Number of examples per label (some have more than others).
- Generate fake data.
Here is an example of a published baseline model on this problem. It's not required to be familiar with the approach used in the paper but, it's good practice to try to read papers like these.
NOTE: The LeNet-5 implementation shown in the classroom at the end of the CNN lesson is a solid starting point. You'll have to change the number of classes and possibly the preprocessing, but aside from that it's plug and play!
Implementation¶
Use the code cell (or multiple code cells, if necessary) to implement the first step of your project. Once you have completed your implementation and are satisfied with the results, be sure to thoroughly answer the questions that follow.
### Preprocess the data here.
### Feel free to use as many code cells as needed.
from sklearn.utils import shuffle
x_train, y_train = shuffle(x_train, y_train)
Question 1¶
Describe how you preprocessed the data. Why did you choose that technique?
Answer: The datatasets consist of 32x32 rgb images. I normalized the samples to the range of -0.5 and 0.5.Normalization is important in CNNs because of their learning process. CNNs learn by continually adding gradient error vectors computed from backpropagation to various weight matrices If I dont scale the input training vectors the ranges of our distributions of feature values would be different for each feature,resulting in corrections in each differing dimension.
While converting the images into gray scale and normalizing their histogram may sound like a good filtering strategy, I think that the images work best in their RGB format. This is very significant in real life scenarios where traffic-signals have to be identified when other objects are present in the background or on the street.
### Generate data additional data (OPTIONAL!)
### and split the data into training/validation/testing sets here.
### Feel free to use as many code cells as needed.
#I splitted the data in part one.
Question 2¶
Describe how you set up the training, validation and testing data for your model. Optional: If you generated additional data, how did you generate the data? Why did you generate the data? What are the differences in the new dataset (with generated data) from the original dataset?
Answer: I've used a split function from sklearn library to split data into training and validation 80%-20%. Then I've randomized the samples.
### Define your architecture here.
### Feel free to use as many code cells as needed.
import tensorflow as tf
EPOCHS = 80
BATCH_SIZE = 150
#Define methods
from tensorflow.contrib.layers import flatten
mu = 0
sigma = 0.1
n_classes=43
layer_depth = {
'layer_1': 6,
'layer_2': 16,
'fully_connected_1': 120,
'fully_connected_2': 84,
'fully_connected_logits': 43
}
# Store layers weight & bias
weights = {
'layer_1': tf.Variable(tf.truncated_normal(
[5, 5, 3, layer_depth['layer_1']],mean = mu, stddev = sigma)),
'layer_2': tf.Variable(tf.truncated_normal(
[5, 5, layer_depth['layer_1'], layer_depth['layer_2']],mean = mu, stddev = sigma)),
'fully_connected_1': tf.Variable(tf.truncated_normal(
[400,layer_depth['fully_connected_1']],mean = mu, stddev = sigma)),
'fully_connected_2': tf.Variable(tf.truncated_normal(
[layer_depth['fully_connected_1'],layer_depth['fully_connected_2']],mean = mu, stddev = sigma)),
'out': tf.Variable(tf.truncated_normal(
[layer_depth['fully_connected_2'], n_classes],mean = mu, stddev = sigma))
}
biases = {
'layer_1': tf.Variable(tf.zeros(layer_depth['layer_1'])),
'layer_2': tf.Variable(tf.zeros(layer_depth['layer_2'])),
'fully_connected_1': tf.Variable(tf.zeros(layer_depth['fully_connected_1'])),
'fully_connected_2': tf.Variable(tf.zeros(layer_depth['fully_connected_2'])),
'out': tf.Variable(tf.zeros(n_classes))
}
def conv2d(x, W, b, strides=1):
x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='VALID')
x = tf.nn.bias_add(x, b)
return x
#return tf.nn.tanh(x)
def maxpool2d(x, k=2):
return tf.nn.max_pool(
x,
ksize=[1, k, k, 1],
strides=[1, k, k, 1],
padding='SAME')
Question 3¶
What does your final architecture look like? (Type of model, layers, sizes, connectivity, etc.) For reference on how to build a deep neural network using TensorFlow, see Deep Neural Network in TensorFlow from the classroom.
Answer: The overal archetecture is a cnn that consists of 2 layers with depth of 6 and 16 respectivly. Then two fully connected layers with a depth of 120 and 84 respectivly. The last connected logits layer classified the results into 43 classes using Adamoptimizer.
The detailed archetecture includes Conv -> ACT(Relu) -> maxPool -> Conv -> ACT(Relu) -> maxPool -> flatten -> Fully Connected -> ACT(ReLu) -> Dropout-> Fully Connected -> ACT(ReLu) -> Dropout -> Fully Connected (Logits)
### Train your model here.
### Feel free to use as many code cells as needed.
def LeNet(x):
# Hyperparameters
mu = 0
sigma = 0.1
dropout=0.55
# TODO: Layer 1: Convolutional. Input = 32x32x1. Output = 28x28x6.
conv1 = conv2d(x, weights['layer_1'], biases['layer_1'] )
# TODO: Activation.
conv1 = tf.nn.relu(conv1)
# TODO: Pooling. Input = 28x28x6. Output = 14x14x6.
conv1=maxpool2d(conv1)
# TODO: Layer 2: Convolutional. Output = 10x10x16.
conv2 = conv2d(conv1, weights['layer_2'], biases['layer_2'] )
# TODO: Activation.
conv2 = tf.nn.relu(conv2)
# TODO: Pooling. Input = 10x10x16. Output = 5x5x16.
conv2=maxpool2d(conv2)
# TODO: Flatten. Input = 5x5x16. Output = 400.
fc1 = flatten(conv2)
# TODO: Layer 3: Fully Connected. Input = 400. Output = 120.
fc1 = tf.add(
tf.matmul(fc1, weights['fully_connected_1']),
biases['fully_connected_1'])
#fc1 = tf.nn.tanh(fc1)
# TODO: Activation.
fc1 = tf.nn.relu(fc1)
# Drop_out
fc1 = tf.nn.dropout(fc1, dropout)
# TODO: Layer 4: Fully Connected. Input = 120. Output = 84.
fc2 = tf.add(
tf.matmul(fc1, weights['fully_connected_2']),
biases['fully_connected_2'])
#fc2 = tf.nn.tanh(fc2)
# TODO: Activation.
fc2 = tf.nn.relu(fc2)
# Drop_out
fc2 = tf.nn.dropout(fc2, dropout)
# TODO: Layer 5: Fully Connected. Input = 84. Output = 10.
logits = tf.add(
tf.matmul(fc2, weights['out']),
biases['out'])
#logits = tf.nn.tanh(logits)
return logits
#Place holder for patches, which is initialized to none in order to accept any patch size
#y stores our labels who come with sparse value
x = tf.placeholder(tf.float32, (None, 32, 32, 3))
y = tf.placeholder(tf.int32, (None))
#one hot encode the labels
one_hot_y = tf.one_hot(y, 43)
Question 4¶
How did you train your model? (Type of optimizer, batch size, epochs, hyperparameters, etc.)
Answer: I've setup 100 epochs (thanks to Amazon cloud) and with a patch size of 128 samples. For optimization I used Adam optimizer to minimize the loss. I kept the rate at 0.001 as I empirically tested several rates and found out that this is a good option. I've also used a drop out of 0.55.
#The pipeline is as follows
#how queckly to update the network weights
rate = 0.001
logits = LeNet(x)
###
prediction = tf.nn.softmax(logits)#here i use softmax, later on i will feed the new images using this
cross_entropy = -tf.reduce_sum(one_hot_y * tf.log(prediction + 1e-6), reduction_indices=1)
loss = tf.reduce_mean(cross_entropy)
###
#tis line will be used for test data without labels
#prediction = tf.nn.softmax(logits)
#train the logits and compare them to the ground truth hot encoded labels
#cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits, one_hot_y)
#Cross-entropy avereging from all images
#loss_operation = tf.reduce_mean(cross_entropy)
#Adam Optimizer to mimimize the loss
optimizer = tf.train.AdamOptimizer(learning_rate = rate)
#path propagation to update the network
#training_operation = optimizer.minimize(loss_operation)
training_operation = optimizer.minimize(loss)
Question 5¶
What approach did you take in coming up with a solution to this problem? It may have been a process of trial and error, in which case, outline the steps you took to get to the final solution and why you chose those steps. Perhaps your solution involved an already well known implementation or architecture. In this case, discuss why you think this is suitable for the current problem.
Answer:In my solution I used the LaNet architecture with some improvements. I have used pooling and drop out to improve the accuracy of the model and lower the overfitting risk. The LaNet is well known convolutional network architecture that has been successfully used for image classification. I've decided the number of epoch after I trained the network with 150 epoches and observed when the network converged. I recommend training with atleast 48 epoches because thats where the network semme to converge. Adding more epoches increases the risk of overfitting. For tuning the hyper parameters I used my intuition. However, a lot of testing and tweaking were required.
#This section of the code was adopted form the SDC course, traffic sign recognition
#Compare the logit prediction to the one-hot encoded ground truth label
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(one_hot_y, 1))
#Average individual prediction accuraccies to get the overall accuracy
accuracy_operation = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
# create patches and run them
def evaluate(x_data, y_data):
num_examples = len(x_data)
total_accuracy = 0
sess = tf.get_default_session()
for offset in range(0, num_examples, BATCH_SIZE):
batch_x, batch_y = x_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
accuracy = sess.run(accuracy_operation, feed_dict={x: batch_x, y: batch_y})
total_accuracy += (accuracy * len(batch_x))
return total_accuracy / num_examples
#This section of the code was adopted form the SDC course, traffic sign recognition
with tf.Session() as sess:
#sess.run(tf.global_variables_initializer())
sess.run(tf.initialize_all_variables())
num_examples = len(x_train)
print("Training...")
print()
for i in range(EPOCHS):
x_train, y_train = shuffle(x_train, y_train)
for offset in range(0, num_examples, BATCH_SIZE):
end = offset + BATCH_SIZE
batch_x, batch_y = x_train[offset:end], y_train[offset:end]
sess.run(training_operation, feed_dict={x: batch_x, y: batch_y})
validation_accuracy = evaluate(x_validation, y_validation)
print("EPOCH {} ...".format(i+1))
print("Validation Accuracy = {:.3f}".format(validation_accuracy))
print()
try:
saver
except NameError:
saver = tf.train.Saver()
saver.save(sess, 'lenet')
print("Model saved")
#Test the performance of the model on the 20% of traffic signs that we have splitted
with tf.Session() as sess:
loader = tf.train.import_meta_graph('lenet.meta')
loader.restore(sess, tf.train.latest_checkpoint('./'))
test_accuracy = evaluate(x_test, y_test)
len_x_test=len(x_test)
prediction=test_accuracy * len_x_test
print("Test Accuracy = {:.3f}".format(test_accuracy))
Step 3: Test a Model on New Images¶
Take several pictures of traffic signs that you find on the web or around you (at least five), and run them through your classifier on your computer to produce example results. The classifier might not recognize some local signs but it could prove interesting nonetheless.
You may find signnames.csv useful as it contains mappings from the class id (integer) to the actual sign name.
Implementation¶
Use the code cell (or multiple code cells, if necessary) to implement the first step of your project. Once you have completed your implementation and are satisfied with the results, be sure to thoroughly answer the questions that follow.
import cv2
import matplotlib.pyplot as plt
import numpy as np
### Load the images and plot them here.
### Feel free to use as many code cells as needed.
real_imgs = []
for i in range(1, 8):
print(i)
img=cv2.imread('image_{}.jpg'.format(str(i)))
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
img = img / 255 - 0.5
resize_img = cv2.resize(img, (32, 32))
resize_img = np.reshape(np.asarray(resize_img), [32, 32, 3])
real_imgs.append(resize_img)
plt.show()
real_imgs = np.asarray(real_imgs)
#real_imgs = real_imgs.astype('float32')
#real_imgs = real_imgs / 255 - 0.5
print(real_imgs.shape)
Question 6¶
Choose five candidate images of traffic signs and provide them in the report. Are there any particular qualities of the image(s) that might make classification difficult? It could be helpful to plot the images in the notebook.
Answer: Images were downloaded from google image search. Some images were too big to be resized and others had background.
### Run the predictions here.
### Feel free to use as many code cells as needed.
#new_prediction=tf.placeholder(tf.float32,(5,32,32,3) )
#logits=LeNet(initial_real_presictions)
import tensorflow as tf
with tf.Session() as sess:
#sess.run(tf.initialize_all_variables())
loader_2 = tf.train.import_meta_graph('lenet.meta')
loader_2.restore(sess, tf.train.latest_checkpoint('./'))
#new_predictions = sess.run(prediction, feed_dict={x: real_imgs})
new_prediction = sess.run(prediction, feed_dict={x: real_imgs})
print(new_prediction)
Question 7¶
Is your model able to perform equally well on captured pictures when compared to testing on the dataset? The simplest way to do this check the accuracy of the predictions. For example, if the model predicted 1 out of 5 signs correctly, it's 20% accurate.
NOTE: You could check the accuracy manually by using signnames.csv (same directory). This file has a mapping from the class id (0-42) to the corresponding sign name. So, you could take the class id the model outputs, lookup the name in signnames.csv and see if it matches the sign from the image.
Answer: No, it wasn't able to perform equally well. 1 out of the 7 signs was correctly predicted. This is about 14% of prediction accuracy. The prediction accuracy from using the validation sets was 92%. I'll explain why I think the prediction accuracy was very low on the captured pictures in the section below. The 92% prediction accuracy on the validation datasets is withing the average of what many other udacity students achieved. However, I think it's impractical because the datasets samples are limited. Each image in the datasets is repeated up to 60 times and slightly zoomed in. The effect of that is high validation accuracy and low performance in the wild.
### Visualize the softmax probabilities here.
### Feel free to use as many code cells as needed.
max_value = np.argmax(new_prediction[0])
print(max_value)
max_value = np.argmax(new_prediction[1])
print(max_value)
max_value = np.argmax(new_prediction[2])
print(max_value)
max_value = np.argmax(new_prediction[3])
print(max_value)
max_value = np.argmax(new_prediction[4])
print(max_value)
max_value = np.argmax(new_prediction[5])
print(max_value)
max_value = np.argmax(new_prediction[6])
print(max_value)
Question 8¶
Use the model's softmax probabilities to visualize the certainty of its predictions, tf.nn.top_k could prove helpful here. Which predictions is the model certain of? Uncertain? If the model was incorrect in its initial prediction, does the correct prediction appear in the top k? (k should be 5 at most)
tf.nn.top_k will return the values and indices (class ids) of the top k predictions. So if k=3, for each sign, it'll return the 3 largest probabilities (out of a possible 43) and the correspoding class ids.
Take this numpy array as an example:
# (5, 6) array
a = np.array([[ 0.24879643, 0.07032244, 0.12641572, 0.34763842, 0.07893497,
0.12789202],
[ 0.28086119, 0.27569815, 0.08594638, 0.0178669 , 0.18063401,
0.15899337],
[ 0.26076848, 0.23664738, 0.08020603, 0.07001922, 0.1134371 ,
0.23892179],
[ 0.11943333, 0.29198961, 0.02605103, 0.26234032, 0.1351348 ,
0.16505091],
[ 0.09561176, 0.34396535, 0.0643941 , 0.16240774, 0.24206137,
0.09155967]])Running it through sess.run(tf.nn.top_k(tf.constant(a), k=3)) produces:
TopKV2(values=array([[ 0.34763842, 0.24879643, 0.12789202],
[ 0.28086119, 0.27569815, 0.18063401],
[ 0.26076848, 0.23892179, 0.23664738],
[ 0.29198961, 0.26234032, 0.16505091],
[ 0.34396535, 0.24206137, 0.16240774]]), indices=array([[3, 0, 5],
[0, 1, 4],
[0, 5, 1],
[1, 3, 5],
[1, 4, 3]], dtype=int32))Looking just at the first row we get [ 0.34763842, 0.24879643, 0.12789202], you can confirm these are the 3 largest probabilities in a. You'll also notice [3, 0, 5] are the corresponding indices.
with tf.Session() as sess:
#probabilities = tf.nn.top_k(new_predictions, k=5)
#values, indices = sess.run(probabilities)
values, indices = sess.run(tf.nn.top_k(new_prediction, 5))
for i in range(0, 5):
for j in range(0, 7):
print(sign_names[str(indices[j][i])], values[j][i])
print("\n")
#sess.run(tf.initialize_all_variables())
#print(sess.run(tf.nn.top_k(new_predictions, 5), feed_dict={x: real_imgs}))
Answer: The prediction accuracy rate on the web acquired images that is shown in the top 7 prediction results is 14%. This is very low compared to the 92% on the validation images. The prediction shows that priority road was accurately predicted with a high certainty. However, the other signs weren't predicted. This very low prediction could likely be attributed to the image resizing (32 x 32) effect, as well as the watermarking on some of the images. For improving the prediction I suggest to collect samples with a higher resolution and build a CNN that can accept larger size of images.
Note: Once you have completed all of the code implementations and successfully answered each question above, you may finalize your work by exporting the iPython Notebook as an HTML document. You can do this by using the menu above and navigating to \n", "File -> Download as -> HTML (.html). Include the finished document along with this notebook as your submission.